Measuring the Depth of LLM Unlearning via Activation Patching 文章

ArXiv CS.CL2026-05-26NEWSen作者: Jaeung Lee, Dohyun Kim, Jaemin Jo

查看原文 →

Measuring the Depth of LLM Unlearning via Activation Patching · 相关技术

相关技术

LLM gin ODE language model safety SAFE Referring expression comprehension (REC)Parts-of-Speech (POS) tags OWL MIT MEC LMM HIS For Effort Metric Attention DiT DAPT Camouflaged object detection ARG ANN TPU